Southern Illinois University CarbondaleOpenSIUC

Research Papers Graduate School

2015

Determinants of College Enrollment in the State ofIllinoisDiana Donnellydonnellydianam@siu.edu

Follow this and additional works at: http://opensiuc.lib.siu.edu/gs_rp

This Article is brought to you for free and open access by the Graduate School at OpenSIUC. It has been accepted for inclusion in Research Papers byan authorized administrator of OpenSIUC. For more information, please contact opensiuc@lib.siu.edu.

Recommended CitationDonnelly, Diana. "Determinants of College Enrollment in the State of Illinois." ( Jan 2015).

DETERMINANTS OF COLLEGE ENROLLMENT IN THE STATE OF ILLINOIS

by

Diana M. Donnelly

B.A., Southern Illinois University, 2014

A Research Paper Submitted in Partial Fulfillment of the Requirements for the

Master of Arts Degree.

Department of Economics

in the Graduate School Southern Illinois University Carbondale

December 2015

RESEARCH PAPER APPROVAL

DETERMINANTS OF COLLEGE ENROLLMENT IN THE STATE OF ILLINOIS

By

Diana M Donnelly

A Research Paper Submitted in Partial

Fulfillment of the Requirements

for the Degree of

Master of Arts

in the field of Economics

Approved by:

Dr. Subhash C. Sharma, Chair

Dr. Scott D. Gilbert

Graduate School Southern Illinois University Carbondale

November 13, 2015

iii

AN ABSTRACT OF THE RESEARCH PAPER OF

Diana M Donnelly, for the Master of Arts degree in Economics, at Southern Illinois University Carbondale. TITLE: DETERMINANTS OF COLLEGE ENROLLMENT IN THE STATE OF ILLINOIS MAJOR PROFESSOR: Dr. Scott Gilbert This paper analyses the effect of various cost, test achievement, and admission

standard variables on the total undergraduate degree-seeking enrollment at 42 four-

year institutions in the State of Illinois. Ordinary Least Squares is used to estimate the

effect of changes in these various types of variables and to determine how schools can

attempt to increase their enrollment numbers by making changes to the various

variables included that are within their administrative control. This study finds that

increasing average monthly faculty salary and decreasing the average SAT score of

students will have the greatest positive effects on enrollment.

iv

TABLE OF CONTENTS

ABSTRACT ................................................................................................................. iv

LIST OF TABLES ......................................................................................................... v

LIST OF FIGURES ....................................................................................................... vi

INTRODUCTION ......................................................................................................... 1

LITERATURE REVIEW ............................................................................................... 2

METHODS AND DATA ............................................................................................... 5

RESULTS .................................................................................................................... 14

SUMMARY AND CONCLUSIONS .............................................................................. 19

REFERENCES ............................................................................................................ 21

VITA ............................................................................................................................ 22

v

LIST OF TABLES

TABLE PAGE

Table 1 – Institutions ...................................................................................................... 7

Table 2 – Descriptive Statistics ....................................................................................... 8

Table 3 – Equation 1 Results ........................................................................................ 15

Table 4 – Equation 2 Results ........................................................................................ 16

vi

LIST OF FIGURES

FIGURE PAGE

Figure 1 – UGDS Trends 1 .............................................................................................. 9

Figure 2 – UGDS Trends 2 .............................................................................................. 9

Figure 3 – UGDS Trends 3 ............................................................................................ 10

1

I. Introduction

Declining postsecondary education enrollment in the United States has recently

become a concern for many colleges and universities. After a long period of increasing

postsecondary education enrollment, the United States’ combined public and private

school systems experienced noticeable declines in the total number of students. The

National Center for Education Statistics reports that, “While total undergraduate

enrollment increased by 37 percent between 2000 and 2010, enrollment in 2013 was 3

percent lower than in 2010,” (National Center for Education Statistics 2015). Similarly,

the United States Census Bureau, which has been collecting data on college enrollment

since 1966, finds that, “College enrollment declined by close to half a million (463,000)

between 2012 and 2013, marking the second year in a row that a drop of this magnitude

has occurred,” (2013). The United States Census Bureau also finds that, “The

cumulative two-year drop of 930,000 was larger than any college enrollment drop before

the recent recession…,” (2013). Of the varying levels and cost structures that make up

the US postsecondary education system, the United States Census Bureau finds the

most substantial fall in enrollment to be in two-year or junior colleges, while four-year

colleges actually saw a slight increase (2013). The United States Census Bureau also

notes that, “‘The drop-off in total college enrollment the last two years follows a period of

expansion: between 2006 and 2011, college enrollment grew by 3.2 million,’ said Kurt

Bauman, chief of the Census Bureau’s Education and Social Stratification Branch. ‘This

level of growth exceeded the total enrollment increase of the previous 10 years

combined (2.0 million from 1996 to 2006),’” (2013). Ronald A. Wirtz, in an article in the

Fedgazette, acknowledges that the postsecondary enrollment declines have taken

2

place amidst rising encouragement to achieve a higher level of education, as well as,

documented financial benefits received by those who have obtained higher levels of

education, and he mentions that part of the reason for the declining enrollment is the

cyclical economy pulling people out of the education system and into the recovering job

market (2015).

The main objective of this research paper is to determine what factors within the

postsecondary institutions’ control are the most influential on the institution’s enrollment.

Amidst the falling enrollment across the US and in the state of Illinois, it is important for

the institutions to be aware of their most effective options in attracting and maintaining

students. This particular study exams the enrollment of 42 four-year predominantly

bachelor’s degree-granting colleges and universities in the state of Illinois for the 2009-

10 through 2013-14 school years and how various factors that are within the institutions

control impact the enrollment of the institution while controlling for the state of the

economy using a variable for the county unemployment rate. This study adds to the

existing literature by using newly available data accounting for numerous variables that

have not previously been examined in relation to postsecondary enrollment.

II. Literature Review

There is an abundance of research that attempts to explain college enrollment

numbers, but none that approach the question in the manner that this research does.

The two past studies that are most applicable to this research are “The Demand for

Higher Education in the United States, 1919-1964” by Robert Campbell and Barry N.

Siegel published in the American Economic Review in 1967 and “Determinants and

3

Distributional Aspects of Enrollment in U.S. Higher Education” by Arthur J Corazzini,

Dennis J. Dugan, and Henry G. Grabowski published in the Journal of Human

Resources in 1972.

Campbell and Siegel’s “The Demand for Higher Education in the United States,

1919-1964” studies the changes in the aggregate demand for higher education in the

United States post-World War I. Their dependent variable is the ratio of undergraduate

degree enrollment in four-year institutions to the number of eligible 18-24 year olds for a

given year; an eligible 18-24 year old is determined to be an individual within the age

range that has a high school diploma and is not a member of the armed forces. The

independent variables of the model are the average real tuition and the real disposable

income per household for a given year. They hypothesize that there will be a negative

price effect, that demand for education will decline as the average real tuition rises, and

a positive income effect, that demand for education will rise as the real income per

household rises. Campbell and Siegel run a log-log model and find results consistent

with their hypotheses. However, although the years analyzed span over a large time

frame, Campbell and Siegel only have nine years of analysis because of lack of data

availability. The minimal observations lead to large standard errors, which makes it hard

to come to conclusive results. Even though this research leaves room for error, it sheds

light on the possible relationships between enrollment determinants and the demand for

higher education.

Corazzini, Dugan, and Grabowski’s “Determinants and Distributional Aspects of

Enrollment in U.S. Higher Education” examines the effect of both demand-side factors

and supply-side constraints on the percentage of tenth grade high school students in a

4

particular state in 1960 that enrolled in college in 1963. The dependent variables

included in their model are state average tuition rates for junior colleges, public four-

year universities, private four-year universities, and teacher colleges; the average

earnings of production workers in the state; the state unemployment rate; the average

level of father’s education in the state; and the student’s performance on achievement

tests. Similar to Campbell and Siegel’s study, the average tuition rates are expected to

have a negative effect on enrollment percentage. The average earnings of production

workers in the state is the wage forgone or the opportunity cost of attending college and

is expected to have a negative relationship with the enrollment percentage. The

unemployment rate is expected to have a positive relationship with the enrollment

percentage because a higher unemployment rate signals difficulty in finding a job if the

labor market is chosen over further education. The father’s average education variable

is believed to be positively correlated with family income and the ability of the eligible

individual to finance college and is, therefore, expected to have a positive effect. Lastly,

students’ performance on achievement exams are supposed to be representative of

students’ ability to be accepted into college, as well as, an indication of their preference

for education and is expected to have a positive relationship. Corazzini, Dugan, and

Grabowski also stratify their analysis to account for the effects of the independent

variables on various socioeconomic groups, but the stratified groups are not of interest

to the study that will be presented here. The general enrollment function yields negative

relationships between all of the tuition averages and the enrollment rate, a positive

relationship between the average production wage and the enrollment rate, a positive

relationship between the unemployment rate and the enrollment rate, a positive

5

relationship between the average father’s education level and the enrollment rate, and a

positive relationship between the students’ performance on achievement tests and the

enrollment rate. All of the relationships were statistically significant at the five percent

level or above, with the exception of the average tuition for teacher colleges and the

average wage for production workers.

This study incorporates variables for tuition, the unemployment rate, and for

student performance on achievement tests, but it also introduces newly available

variables into the model. Also, tuition and achievement test scores are discussed from

the perspective of the supplier of education, or the institution, and how they can attempt

to manipulate these various variables in order to attain higher enrollment numbers or

increase the demand for their institution. The unemployment rate is included as a

control for the state of the economy and is not addressed as a variable that the

institution has any control over. Other important differences of this study are that it

analyses the actual enrollment number rather than the enrollment as a portion or ratio of

the eligible population, and it analyses the dependent variable separately for each

institution studied rather than the aggregate enrollment of all.

III. Methods and Data

Data for this analysis is collected from two sources. The College Scorecard Data

put out by the United States Department of Labor is used to obtain information on

the total enrollment of undergraduate degree seeking students at the institution (UGDS),

the in state tuition and fees of the institution (TUITIONFEE_IN), the out of state tuition

and fees of the institution (TUITIONFEE_OUT), the midpoint of the ACT English score

6

(ACTENMID), the midpoint of the ACT math score (ACTMTMID), the average SAT

score (SAT_AVG), the median debt of students that have completed their degree at the

institution (GRAD_DEBT_MDN), whether the institution is public or private (CONTROL),

the admission rate of the institution (ADM_RATE), the retention rate of first-time full time

students at the institution (RET_FT4), the average faculty salary per month at the

institution (AVGFACSAL), and the rate of students that completed their degree within

six years (C150_4). These variables are collected for each institution in the sample for

five consecutive years, 2009 through 2013, and are all based on the fall semester that

typically begins in August. Data on the unemployment rate it collected from the Bureau

of Labor Statistics’ “Local Area Unemployment Statistics” Survey that collects

employment data for counties and metropolitan areas. This study utilizes the monthly

county unemployment rates for the counties that the institutions are in. The particular

variable used for analysis is the average of the monthly unemployment rates over the

year prior to the beginning of the fall semester (COUNTY_UNEMP), so the value for

COUNTY_UNEMP for 2009 is an average of the September 2008 through August 2009

county unemployment rates for the given county. All of these variables are obtained for

a total of 42 four-year, predominantly bachelor’s degree granting institutions in the state

of Illinois for the years 2009 through 2013. The institutions included in the sample, as

well as, their respective cities, counties, and CONTROL (0 for public, 1 for private) are

listed in Table 1, and the descriptive statistics of the variables are shown in Table 2.

Descriptive statistics are all rounded to the nearest thousandth.

7

Table 1 – Institutions

INSTNM CITY COUNTY CONTROL

Aurora University Aurora Will 1

Blackburn College Carlinville Macoupin 1

Bradley University Peoria Peoria 1

Chicago State University Chicago Cook 0

Concordia University-Chicago River Forest Cook 1

DePaul University Chicago Cook 1

Dominican University River Forest Cook 1

Eastern Illinois University Charleston Coles 0

Elmhurst College Elmhurst DuPage 1

Eureka College Eureka Woodford 1

Illinois Institute of Technology Chicago Cook 1

Illinois State University Normal McLean 0

Illinois Wesleyan University Bloomington McLean 1

Lewis University Romeoville Will 1

Lincoln Christian University Lincoln Logan 1

Loyola University Chicago Chicago Cook 1

McKendree University Lebanon St Claire 1

Millikin University Decatur Macon 1

Monmouth College Monmouth Warren 1

North Central College Naperville Will 1

North Park University Chicago Cook 1

Northeastern Illinois University Chicago Cook 0

Northern Illinois University Dekalb Dekalb 0

Northwestern University Evanston Cook 1

Olivet Nazarene University Bourbonnais Kankakee 1

Quincy University Quincy Adams 1

Rockford University Rockford Winnebago 1

Roosevelt University Chicago Cook 1

Saint Xavier University Chicago Cook 1

School of the Art Institute of Chicago Chicago Cook 1

Southern Illinois University Carbondale Carbondale Jackson 0

Southern Illinois University Edwardsville Edwardsville Madison 0

Trinity Christian College Palos Heights Cook 1

Trinity International University-Illinois Deerfield Lake 1

University of Chicago Chicago Cook 1

University of Illinois at Chicago Chicago Cook 0

University of Illinois at Springfield Springfield Sangamon 0

University of Illinois at Urbana-Champaign Champaign Champaign 0

University of St Francis Joliet Will 1

VanderCook College of Music Chicago Cook 1

Western Illinois University Macomb McDonough 0

Wheaton College Wheaton DuPage 1

8

Table 2 – Descriptive Statistics

Variable Observations Mean Std. Deviation Min. Max.

UGDS 210 5762.324 6443.905 104.000 31663.000

TUITIONFEE_IN 210 22906.260 9486.045 7082.000 47514.000

TUITIONFEE_OUT 210 25373.250 7166.636 12962.000 47514.000

ACTENMID 210 23.886 3.317 17.000 34.000

ACTMTMID 210 23.252 3.298 16.000 34.000

SAT_AVG 210 1088.138 123.474 830.000 1504.000

GRAD_DEBT_MDN 210 21575.460 3313.851 12500.000 28000.500

CONTROL 210 0.738 0.441 0.000 1.000

ADM_RATE 210 0.635 0.163 0.088 0.981

RET_FT4 210 0.768 0.104 0.484 0.993

AVGFACSAL 210 7490.524 2068.526 3992.000 16589.000

C150_4 210 0.588 0.160 0.139 0.952

COUNTY_UNEMP 210 9.320 1.488 6.100 15.000

Of the 42 schools in the sample, there are eleven public four-year universities

and 31 private four-year universities. The average UGDS for the 210 observations is

5,762.324 students with a standard deviation of plus or minus 6,443.905 students. The

large standard deviation is likely a result of the University of Illinois at Urbana-

Champaign’s large enrollment observations, which can be seen in Figure 1. Excluding

the institution for being an outlier was considered, but the variables are eventually

transformed into logarithmic form, which minimizes the range of differences drastically.

However, Figures 1-3 of the UGDS trends for all of the institutions in the sample is

shown to illustrate the different levels of enrollment that exist across the sample and to

show the changes in each institution’s enrollment over the years that are observed for

the analysis. As the figures depict, not all institutions in the sample show the same

trend. From just a quick glance one can see that some institutions show slight increases

in their enrollment over the five-year span, some show declines that appear to begin

around 2011, and others do not appear to have any noticeable change.

9

Figure 1 – UGDS Trends 1

Figure 2 – UGDS Trends 2

30000

30500

31000

31500

32000

2009 2010 2011 2012 2013

UG

DS

Years

UGDS Trends 1

University of Illinois at…

6000

8000

10000

12000

14000

16000

18000

20000

2009 2010 2011 2012 2013

UG

DS

Years

UGDS Trends 2

DePaul University Estern Illinois University

Elmhurst College Illinois State University

Loyola University Chicago Northeastern Illinois University

Northern Illinois University Northwestern University

Southern Illinois University Carbondale Southern Illinois University Edwardsville

University of Illinois at Chicago Western Illinois University

10

Figure 3 – UGDS Trends 3

0

1000

2000

3000

4000

5000

6000

2009 2010 2011 2012 2013

UG

DS

Years

UGDS Trends 3

Aurora University Blackburn College

Bradley University Chicago State University

Concordia University-Chicago Dominican University

Eureka College Illinois Institute of Technology

Illinois Wesleyan University Lewis University

Lincoln Christian University McKendree University

Millikin University Monmouth College

North Central College North Park University

Olivet Nazarene University Quincy University

Rockford University Roosevelt University

Saint Xavier University School of the Art Institute of Chicago

Trinity Christian College Trinity International University-Illinois

University of Chicago University of Illinois at Springfield

University of St Francis VanderCook College of Music

Wheaton College

11

The dependent variable, UGDS, is regressed on the independent variables using

a linear regression model to determine the direction and significance levels of the

relationship between each independent variable and the dependent variable. The basic

linear equation is then:

UGDSit = β0 + β1 (TUITIONFEE_INit) + β2 (TUITIONFEE_OUTit) + β3 (ACTENMIDit) +

β4 (ACTMTMIDit) + β5 (SAT_AVGit) + β6 (GRAD_DEBT_MDNit) + β7 (CONTROLi) +

β8 (ADM_RATEit) + β9 (RET_FT4it) + β10 (AVGFACSALit) + β11 (C150_4it) +

β12 (COUNTY_UNEMPit) + εit (1),

where UGDSit is the total enrollment of undergraduate degree seeking students at

institution i in year t, TUITIONFEE_INit is the in state tuition and fees at institution i in

year t, TUITIONFEE_OUTit is the out of state tuition and fees at institution i in year t,

ACTENMIDit is the midpoint of the ACT English scores at institution i in year t,

ACTMTMIDit is the midpoint of the ACT math scores at institution i in year t, SAT_AVGit

is the average SAT score at institution i in year t, GRAD_DEBT_MDNit is the median

amount of debt accumulated by graduates that complete their bachelor’s degree at

institution i in year t, CONTROLi is a dummy variable that represents whether institution

i is public or private, ADM_RATEit is the percentage of admitted students out of all

applicants at institution i in year t, RET_FT4it is the retention rate of first-time full time

students at institution i in year t, AVGFACSALit is the average monthly faculty salary at

institution i in year t, C150_4it is the percentage of students that completed their degree

within six years at institution i in year t, COUNTY_UNEMPit is the average monthly

12

unemployment rate for the county that institution i is in over the twelve months prior to

the start of the fall semester for the year t, and εit is the error term. Similar to the

previous studies that have been conducted, the cost variables, TUITIONFEE_IN and

TUITIONFEE_OUT, are expected to have a negative effect on enrollment because, in

general, the more a good or service costs the less demand there is for it. All of the test

score variables, ACTENMID, ACTMTMID, and SAT_AVG, are also expected to have a

negative effect on enrollment because, as the midpoint or average of the test score

rises, the number of applicants with test scores meeting the criteria is likely to decline.

GRAD_DEBT_MDN is expected to have a negative effect as well because, similar to

the intuition for the tuition and fee cost variables, a greater amount of debt is likely to

deter people from demanding education services from that institution. The CONTROL

variable is expected to have a negative effect as well since private schools are often

viewed as being more expensive than public schools. The admission rate, ADM_RATE,

is expected to have a positive effect on enrollment because the more students admitted

out of the students that apply means more enrollment, however, institutions could have

a high enrollment and a low admission rate as a result of there being a high demand for

that particular institution or a large number of applicants. Retention rate, RET_FT4, is

expected to have a positive effect because maintaining students over the years means

re-enrollment in the consecutive years. The average monthly faculty salary,

AVGFACSAL, is expected to have a positive effect because a higher salary could be

representative of the faculty’s quality as teachers and their credentials. The completion

rate, C150_4, is likely to have a positive effect because applicants are likely to choose

to attend an institution that they feel will be an efficient use of their time, not one where

13

a large proportion take more than six years to complete a bachelor’s degree, however, a

low completion rate could mean that students are enrolling for more than six years

which would increase enrollment. Lastly, the variable that controls for the state of the

economy, COUNT_UNEMP, is expected to have a negative effect on the enrollment

because high unemployment means less job opportunities, and less job opportunities

makes the job market less attractive and the education service more attractive because

it provides more skills and opportunities within the job market after completion.

The linear regression model specified in equation 1 is estimated and tested for

the presence of heteroscedasticity using the Breusch-Pagan test and the White’s test,

both of which show evidence of the presence of heteroscedasticity in the model. In

order to control for the presence of heteroscedasticity, all of the count variables are log

transformed. The only variable in the model that is not a count variable is the dummy

variable for whether the institution is public or private, the CONTROL variable. Then, the

same model is estimated using the log transformed variables. The second equation is

hence:

log_UGDSit = β0 + β1 (log_TUITIONFEE_INit) + β2 (log_TUITIONFEE_OUTit) + β3

(log_ACTENMIDit) + β4 (log_ACTMTMIDit) + β5 (log_SAT_AVGit) +

β6 (log_GRAD_DEBT_MDNit) + β7 (CONTROLi) + β8 (log_ADM_RATEit) +

β9 (log_RET_FT4it) + β10 (log_AVGFACSALit) + β11 (log_C150_4it) +

β12 (log_COUNTY_UNEMPit) + εit (2),

14

where each of the variables is now the log of the original variable, except for the

CONTROL variable, which has not changed.

IV. – Results

The linear regression model, equation 1, is estimated and the results are

displayed in Table 3. Because this model includes the original count variables, the

results are interpreted as a one unit increase is the independent variable results in an

increase or decrease of the actual value found in the dependent variable; for example, a

one unit increase in the TUITIONFEE_IN variable results in a decrease in UGDS of

.5847, or about a sixth of a student. The tuition and fees variables, the midpoint of the

ACT math score, whether the institution is public or private, the admission rate, the

average faculty salary, and the completion rate are all statistically significant at the five

percent level or above. However, both the TUITIONFEE_OUT and the ACTMTMID

have the opposite sign of what was expected. The midpoint of the ACT English score,

the average SAT score, the county unemployment rate, the median debt of graduates,

and the retention rate are all insignificant, and the median debt of graduates does not

have the expected sign. The R-squared of this model is 0.7208, so this model appears

to explain about 72.1 percent of the variation in the dependent variable, UGDS.

This regression model is then tested for heteroscedasticity, as mentioned in the

previous section, and, because there is evidence of heteroscedasticity, all of the

variables are log transformed, with the exception of the dummy variable for whether the

institution is public or private. After the variables are log transformed, the log-log

regression model, equation 2, is estimated, and the results are displayed in Table 4.

15

Table 3 – Equation 1 Results

Independent

Variables

Dependent Variable

UGDS

TUITIONFEE_IN -.5847***

(.1484)

TUITIONFEE_OUT .4457***

(.1428)

ACTENMID -263.4528

(436.0903)

ACTMTMID 898.4277**

(357.5206)

SAT_AVG -16.8892

(14.3380)

COUNTY_UNEMP 246.2831

(191.4572)

GRAD_DEBT_MDN .0175

(.0892)

CONTROL -4067.737**

(1810.793)

ADM_RATE 5610.731***

(1853.007)

RET_FT4 1754.659

(4784.606)

AVGFACSAL 1.2071***

(.2619)

C150_4 8239.386**

(3606.155)

constant -6837.425

(6054.478)

R-Squared 0.7208

n 210

Note: standard errors in parenthesis *- significant at 10% **- significant at 5% ***- significant at 1%

16

Table 4 – Equation 2 Results

Independent

Variables

Dependent Variable

log_UGDS (1)

Dependent Variable

log_UGDS (2)

log_TUITIONFEE_IN 1.4113***

(.5359)

1.3402***

(.2752)

log_TUITIONFEE_OUT -.1458

(.4895)

---

log_ACTENMID .9536

(1.6376)

---

log_ACTMTMID 1.9046

(1.3356)

---

log_SAT_AVG -6.7147***

(2.4891)

-3.6074***

(.9531)

log_COUNTY_UNEMP .5516**

(.2712)

.5210*

(.2658)

log_GRAD_DEBT_MDN -1.4455***

(.2960)

-1.4779***

(.2948)

CONTROL -2.4914***

(.3926)

-2.4624***

(.2576)

log_ADM_RATE .4424***

(.1419)

.4858***

(.1344)

log_RET_FT4 -1.0478**

(.5189)

-.9094*

(.5094)

log_AVGFACSAL 2.5258***

(.3124)

2.5857***

(.3089)

log_C150_4 .7671***

(.2302)

.7749***

(.2297)

constant 26.4012**

(12.1507)

12.8204*

(6.7667)

R-Squared 0.7750 0.7722

n 210 210

Note: standard errors in parenthesis *- significant at 10% **- significant at 5% ***- significant at 1%

17

In moving from equation 1 to equation 2, there is an increase in model fit and

more of the variables are statistically significant. The R-squared increases from 0.7208

to 0.7722, so the log-log regression is able to account for about an additional five

percent of the variation in enrollment. The average SAT score, the county

unemployment variable, the median amount of debt of graduates, the retention rate, and

the constant are statistically significant in the log-log model and were not statistically

significant in the previous regression model. Also, the midpoint of the ACT math score is

not statistically significant in the log-log model and was statistically significant in the

previous model. Overall, the log-log model is a much better fit for this particular data.

Given that equation 2 is in the log-log model form, the results are interpreted

slightly different than the previous results. The results for log-log models are written in

terms of elasticities or percentages. So looking at column (1) of Table 4, a one percent

increase in TUITIONFEE_IN results in a 1.4113 percent increase in UGDS and is

statistically significant at the one percent level. This is the opposite of what was

expected, and the sign has flipped from the previous regression model. One possible

explanation for this relationship is that students associate a higher price with a more

valuable education. A one percent increase in TUITIONFEE_OUT results in a .1458

decrease is UGDS, a one percent increase in ACTENMID results in a .9536 percent

increase in UGDS, and a one percent increase in ACTMTMID results in a 1.9046

percent increase in UGDS, and none of these three relationships are statistically

significant. It is notable, however, the sign of TUITIONFEE_OUT has also flipped from

what it was in the previous regression model. A one percent increase in SAT_AVG

results in 6.714 percent decrease in UGDS and is statistically significant at the one

18

percent level. A one percent increase in COUNTY_UNEMP results in a .5516 percent

increase in UGDS and is statistically significant at the five percent level. A one percent

increase in GRAD_DEBT_MDN results in a 1.4455 decrease in UGDS and is

statistically significant at the one percent level. Aside from TUITIONFEE_IN having the

opposite sign as expected, this result is puzzling because a higher tuition is likely to also

mean a higher amount of debt for graduates, but a higher amount of debt for graduates

has a negative effect on enrollment, whereas, a higher tuition appears to have a positive

effect on enrollment. Because the CONTROL variable is not logged it is interpreted in a

slightly different manner; private institutions have a UGDS that is 2.914 percent lower

than that of public schools, and this result is statistically significant at the one percent

level. A one percent increase in ADM_RATE results in a .4424 percent increase in

UGDS and is statistically significant at the one percent level. A one percent increase in

RET_FT4 results in a 1.0478 percent decrease in UGDS and is significant at the five

percent level. This is the opposite sign of what was expected, and seems

counterintuitive. A one percent increase in AVGFACSAL results in a 2.5258 percent

increase in UGDS and is statistically significant at the one percent level. A one percent

increase in C150_4 results in a .7671 percent increase in UGDS and is statistically

significant at the one percent level.

After running the complete log-log regression, the independent variables that are

not statistically significant are removed from the model, so TUITIONFEE_OUT,

ACTENMID, and ACTMTMID are removed. The regression is rerun with the remaining

variables, and the results are displayed in column (2) of Table 4. The variables in this

equation can all be interpreted in the same way as the previous equation. From deleting

19

the statistically insignificant variables from the equation, the R-squared decreases from

0.7750 to 0.7722, which illustrates that the excluded variables were explaining very little

variation in the dependent variable. The COUNTY-UNEMP variable goes from being

statistically significant at the five percent level to only being statistically significant at the

ten percent level. Similarly, the RET_FT4 goes from being statistically significant at the

five percent level to only being statistically significant at the ten percent level. None of

the signs have changed, but there are light adjustments in the magnitudes of the effects

of the independent variables on the dependent variable, which is to be expected.

From the second model, equation 2, having removed all of the insignificant

variables, it is possible to analyze which variables will have the largest effects on the

enrollment. This is of importance for institutions that are trying to increase their

enrollment numbers. From the variables included, the ones with the greatest effects are

the average SAT score and the average faculty salary. Based on these findings, two

internal changes, then, that could boost enrollment numbers would be to increase the

pay offered to faculty members and to decrease the standards for SAT scores.

V. Summary and Conclusion

This study utilizes data from the College Scorecard Data that is put out by the

United Stated Department of Education to examine the relationship between various

cost, test achievement, and admission standard variables and the total undergraduate

degree-seeking enrollment at 42 different four-year postsecondary institutions in the

state of Illinois for 2009 through 2013. The relationships are estimated using Ordinary

Least Squares methodology, and the variables are log transformed to get the best

20

possible model fit and to control for the presence of heteroscedasticity. The variables

that appear to have the largest effect on enrollment are the admission rate, which has a

positive effect, and the average SAT score, which has a negative effect. The resulting

relationship between the main cost variable and enrollment has an unexpected sign,

which could be explained by students associating price with the value of the education

received from that institution. However, it is puzzling that the median amount of debt of

graduates has a negative effect on enrollment, whereas, the cost of tuition and fees has

a positive effect on enrollment. More research is needed to look into these particular

variables to explain these simultaneous relationships.

21

REFERENCES

Bureau of Labor Statistics. 1990-2015. “Local Area Unemployment Statistics: Not

Seasonally Adjusted, Counties and equivalents, unemployment rate.” United

States Department of Labor. http://www.bls.gov/lau/ (accessed on October 10,

2015).

Campbell, Robert, and Barry N. Siegel. 1967. "The Demand for Higher Education in the

United States, 1919-1964." American Economic Review 57, no. 3: 482-

494. Business Source Complete, EBSCOhost (accessed September 3, 2015).

College Scorecard Data. 1996-2015. United States Department of Education.

https://collegescorecard.ed.gov/data/ (accessed September 10, 2015).

Corazzini, Arthur J., Dennis J. Dugan, and Henry G. Grabowski. 1972. "Determinants

and Distributional Aspects of Enrollment in U.S. Higher Education." Journal of

Human Resources 7, no. 1: 39-59. Business Source Complete,

EBSCOhost (accessed September 3, 2015).

National Center for Education Statistics. 2015. “Undergraduate Enrollment.” May.

http://nces.ed.gov/programs/coe/indicator_cha.asp

United States Census Bureau. 2014. “College Enrollment Declines for Second Year in a

Row, Census Bureau Reports.” News release. September 24.

http://www.census.gov/newsroom/press-releases/2014/cb14-177.html

Wirtz, Ronald A. 2015. “College enrollments falling in Ninth District, nationwide.”

Fedgazette 15. Business Source Complete, EBSCOhost (accessed November 1,

2015).

22

VITA

Graduate School Southern Illinois University

Diana Donnelly donnellydianam@siu.edu Southern Illinois University Bachelor of Arts, Economics, May 2014 Research Paper Title:

Determinant of College Enrollment in the State of Illinois

Major Professor: Dr. Scott Gilbert